AI is Reshaping Invention – and Patent Rules are Racing to Keep Up

Artificial intelligence is now a game-changer in fields like drug design, materials optimization, and device engineering, allowing innovation to happen at an unprecedented pace. However, a pressing question arises when AI proposes a new molecule or when a simulated design outshines a physical prototype: is this an invention, or merely a simulation? Patent systems worldwide are working towards a practical answer, and the nuances of these developments are crucial for anyone involved in AI-driven research and development.

Who Can Be an Inventor When AI is Involved?

A clear consensus is emerging: inventors must be human. Courts and patent offices have regularly turned down applications that list AI as the sole inventor, especially notable in the DABUS cases:

• United States: The Federal Circuit ruled that an inventor under the Patent Act must be a natural person (Thaler v. Vidal, 2022). In 2024, the USPTO clarified that AI-assisted inventions can be patentable provided a human made a significant contribution to the claimed invention (USPTO Inventorship Guidance, 2024).
• Europe: The EPO rejected applications naming DABUS as the inventor, affirming that a machine cannot be recognized as an inventor (EPO, 2021).
• United Kingdom: The UK Supreme Court reached the same conclusion in 2023, asserting that an inventor under the Patents Act must be a person (Thaler v Comptroller-General, 2023).
• Australia: An initial ruling allowed an AI to be named, but a Full Court later reversed that decision (Thaler v Commissioner of Patents, 2022).
• South Africa: A DABUS patent was granted but was not thoroughly examined, indicating limited acceptance (WIPO Magazine, 2021).

The takeaway? Avoid listing an AI system as an inventor. Instead, document the specific innovative contributions made by human researchers. According to the USPTO guidance, at least one human must have significantly contributed to at least one claim; routine tasks or simple selections from AI outputs typically won’t cut it.

Are Simulations Patentable Inventions?

Often, research led by AI results in simulated outcomes first: for instance, a model predicts the ideal catalyst at a specific surface geometry, or a digital twin illustrates a processor design that operates cooler. Is it possible to patent these advancements when they rely mostly on simulation data?

The answer depends on how the invention is framed and whether there’s a tangible technical contribution:

• Europe: Computer-implemented simulations can be patentable if they effectively address a technical problem. The EPO has stressed that simulations aren’t automatically excluded but must demonstrate a technical effect beyond mere abstract modeling. Refer to the Enlarged Board’s decision on simulations (G 1/19) and earlier rulings recognizing technical contributions in specific instances, like circuit simulations (T 1227/05).
• United States: Claims directed to abstract ideas or mathematical models are not eligible unless integrated into a practical application that enhances technology. It’s essential for applicants to tie claimed features to a tangible technical improvement, not just a more effective algorithm. This principle is detailed in the Supreme Court’s eligibility framework and the USPTO’s subject matter eligibility guidance (USPTO PEG, 2019).

Both systems are inclined to favor filings that articulate how the simulation is used to control or design a physical system, quantify measurable performance gains, and reveal specific technical constraints addressed by the method.

Is Simulated Data Enough for Enablement and Sufficiency?

Patents generally do not mandate a built prototype. What truly counts is whether the specification enables a skilled individual to make and utilize the invention without undue experimentation (in the US) or provides adequate disclosure (in Europe):

• United States: Prophetic examples are permitted if they’re clearly identified and backed by credible reasoning. However, the disclosure must enable the entire scope of the invention claims. The Supreme Court recently reaffirmed this in a biotech context (Amgen v. Sanofi, 2023). See also the USPTO’s guidance on prophetic examples and enablement (MPEP 2164).
• Europe: Data submitted after filing can demonstrate a technical effect if the application made the effect plausible at the time of filing. The Enlarged Board recently clarified this plausibility standard (G 2/21).

In practice, you can submit a patent application based on high-quality simulations, provided the specification outlines a concrete way to implement the invention and gives credibility to the predicted results. Clearly label simulated or prophetic examples, and provide at least one detailed implementation path with parameters, tolerances, and constraints.

AI-Assisted Invention: Practical Steps for R&D Teams

• Document human contributions. Maintain lab notebooks, versioned prompts, model configurations, and selection criteria to illustrate who conceived the claimed features and why.
• Avoid naming AI as an inventor. Identify human contributors who significantly influenced at least one claim (see USPTO 2024 guidance; EPO position).
• Base simulations on technical constraints. Explain the real-world system being modeled, assumptions, boundary conditions, and how the method results in a technical enhancement.
• Ensure a clear build path. Offer sufficient parameters, materials, architectures, or training procedures so that a skilled person can implement it without undue experimentation.
• Label simulated data and prophetic examples effectively. Be explicit about what is simulated or measured; avoid overstating results.
• Validate when possible. Even a small amount of experimental confirmation can enhance credibility and mitigate risks during examination and litigation.
• Consider timing and strategy. If experimental validation is on the horizon, coordinate filings to include it; otherwise, file with robust simulations and follow up Data within procedural rules.
• Be mindful of overlapping regulations. AI transparency or data rules (such as the EU AI Act) don’t impact patentability but may affect documentation and compliance (EU AI Act, 2024).

Common Pitfalls to Avoid

• Avoid claiming broad functional outcomes with limited simulated support.
• Don’t rely on black-box assertions without clarifying model design, training data characteristics, or validation logic.
• Steer clear of describing abstract optimization goals without a concrete technical implementation.
• Don’t under-document the human contributors responsible for the claimed solution.

Bottom Line

AI and simulation have now become integral to the invention process. Patent systems are evolving by emphasizing two key principles: a human inventor with meaningful contributions and a concrete technical teaching that enables real-world application. By clarifying these principles in your applications, simulation-driven R&D can yield strong and enforceable patents.

FAQs

Can I Name an AI System as an Inventor?

No. Major jurisdictions require a human inventor. Refer to the USPTO’s 2024 guidance and decisions from EPO, UK Supreme Court, and US Federal Circuit mentioned above.

Do I Need a Working Prototype to File?

No. An enabling disclosure is sufficient. Simulations and prophetic examples can be acceptable if they credibly teach how to make and utilize the invention.

Are Simulation-Only Inventions Patentable?

Potentially, if the claims and description demonstrate a technical contribution that enables implementation. In Europe, consult G 1/19; in the US, ensure the model is integrated into a practical, technological application.

Do I Need to Disclose My Prompts or Training Data?

Only disclose what’s necessary to enable the invention. There’s no requirement to provide extensive model details unless essential for others to practice the claimed invention.

Could AI-Generated Outputs Be Unpatentable as Mere Discoveries?

It can depend. If the claim pertains to a simple discovery or abstract correlation, it risks being non-eligible. Frame claims around tangible technical implementations and measurable improvements.

Sources

1. USPTO Inventorship Guidance for AI-Assisted Inventions (2024)
2. Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022)
3. EPO press release on DABUS inventorship refusals (2021)
4. Thaler v Comptroller-General of Patents, UK Supreme Court (2023)
5. Thaler v Commissioner of Patents, Full Court of the Federal Court of Australia (2022)
6. WIPO Magazine: South Africa grants a patent listing an AI as inventor (2021)
7. EPO Enlarged Board of Appeal, G 1/19 (Simulation/Aerodynamics)
8. EPO Board of Appeal, T 1227/05 (Circuit Simulation)
9. USPTO Patent Eligibility Guidance (2019 PEG)
10. USPTO MPEP 2164: Enablement and prophetic examples
11. Amgen Inc. v. Sanofi, 598 U.S. 594 (2023)
12. EPO Enlarged Board of Appeal, G 2/21 (Plausibility)
13. EU AI Act press release (2024)